From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756201AbbBFP6Q (ORCPT ); Fri, 6 Feb 2015 10:58:16 -0500 Received: from mail-we0-f177.google.com ([74.125.82.177]:49212 "EHLO mail-we0-f177.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755130AbbBFP6M (ORCPT ); Fri, 6 Feb 2015 10:58:12 -0500 Message-ID: <1423238288.3377.83.camel@gmail.com> Subject: Re: [RFC][PATCH] sched/rt: Use IPI to trigger RT task push migration instead of pulling From: Mike Galbraith To: Peter Zijlstra Cc: Steven Rostedt , LKML , Ingo Molnar , Thomas Gleixner , Clark Williams , linux-rt-users , "Paul E. McKenney" Date: Fri, 06 Feb 2015 16:58:08 +0100 In-Reply-To: <20150206124051.GN21418@twins.programming.kicks-ass.net> References: <20150204143906.742287b4@gandalf.local.home> <20150205145833.GJ5029@twins.programming.kicks-ass.net> <20150205110420.398fb3c8@gandalf.local.home> <20150206124051.GN21418@twins.programming.kicks-ass.net> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.12.9 Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 2015-02-06 at 13:40 +0100, Peter Zijlstra wrote: > On Thu, Feb 05, 2015 at 11:04:20AM -0500, Steven Rostedt wrote: > Well, the thing is, with or without this patch, the worst case is still > the very same. The likelihood of actually triggering it goes down > dramatically [*], but any spin_lock() has that O(nr_cpus) worst case, and > that spinlock op isn't going away. > > Now, lowering avg/median cases is good, but don't kid yourself that the > worst case is going to be better. > > People running -RT on big boxes had better be aware of this. The folks I know of running -RT on big boxen isolate cores and nail absolutely everything to the floor. The only way to survive in a big box is to chop the thing into tiny pieces. A plasma cutter would be even better.. 'course you could just _buy smaller boxen_ and connect what you want connected, take the lazy path to big box solution ;-) -Mike